Extracting Data from the Deep Web with Global-as-View Mediators Using Rule-Enriched Semantic Annotations

Abstract. The Deep Web offers approximately 500 times more information than the Open Web, but is “hidden ” behind search-forms intended for human users, and typically requires interaction, which makes it difficult to index by Web crawlers. We argue that traditional data extraction is therefore not suitable for the Deep Web and suffers from coverage problems similar to those search engines face when trying to index its content. Instead, it is proposed to trans-form and forward queries on demand using Global-as-View Mediators. To al-low automated interaction with databases on the Deep Web, we use rules that exploit features (e.g. HTML attribute values) to identify elements on a Web page and infer semantic annotations that link these elements to known concepts (e.g. query parameters or result values). Using a prototypical implementation, Deep Web Mediator, the performance of this approach is demonstrated in a classified-advertising use case. Our system is able to answer complex queries by transforming and forwarding them to multiple sites as well as integrating the local results